This is the repository of the "Efficient 3D liver segmementation" project for the "CentraleSupelec - Spring 2020 MVA-DLMI: Deep Learning in Medical Imaging" course. In this project we tested a 3D version of Deeplab with a mobile net and a resnet backbone and compare it to a U-net and a attention gated Unet
The group of students was composed of three students :
The report is available on demand.
Python3
- Download the data from the LiTS challenge (train batch 1 and train batch 2)
- Clone this repository
- Run
pip install -r requirements.txt
All the commands are to be executed from the main directory of this repository.
- From the
configs
directory, choose the json file you want to train or create a new one and adapt the datapath to the folder containing the downloaded directoriesTraining Batch 1
andTraining Batch 2
. - To launch a training run
python utils/train.py --config_file=<path to config.json> --logdir=<path to the directory containing all the log dirs>
To evaluate a run use :
python utils/error_analysis.py –run_dir=<path to the log dir>
(the log dir is the one named after the date and the time of the training)
To run a the test time evalutation, run:
python utils/speed_test.py